MKF2121 Chap.3 Questionnaires, Scaling, Sampling, and Experiments
Questionnaires, Scaling, Sampling, and Experiments
Scaling, sampling and experimentation control three different claims: what a response value means, who the observed cases can represent and whether a contrast supports causality. This chapter covers comparative and non-comparative scales, questionnaire sequencing, target populations and frames, probability and non-probability techniques, experimental notation and threats to internal and external validity.
What this chapter covers
- 01
Nominal, ordinal, interval and ratio properties
- 02
Comparative ranking, paired and constant-sum tasks
- 03
Continuous and itemised non-comparative scales
- 04
Question order, routing and response options
- 05
Target population, sampling frame and achieved sample
- 06
Probability sampling and known selection chances
- 07
Convenience, judgement, quota and referral sampling
- 08
Sample-size planning and response quality
- 09
Experimental notation and treatment contrasts
- 10
Internal and external validity threats
Design a stratified service-tier sample
- 2Define mutually exclusive tier membership and a current frame.
- 2Select probabilistically within each tier and oversample the small strategic tier.
- 2Use design weights for population totals and diagnose response separately within tiers.
Key terms
- Comparative scale
- A task evaluating objects directly against one another through choices, rankings or allocations.
- Ratio scale
- A measurement scale with identity, order, equal intervals and a meaningful zero.
- Target population
- The complete set of elements relevant to the research question.
- Probability sample
- A sample selected through known non-zero inclusion chances under the design.
- Random assignment
- Allocation of study units to conditions by chance to protect causal comparison.
- Treatment effect
- The outcome contrast between a treatment condition and a credible counterfactual.
Questionnaires, Scaling, Sampling, and Experiments FAQ
Do numeric codes make a variable metric?
No. Digits can label unordered categories, ordered levels or quantities. The represented property determines the measurement scale and permissible comparisons, so customer IDs or region codes remain nominal even when software stores them as numbers.
Why can a large convenience sample still be biased?
Large size makes percentages stable within the achieved respondent set but does not change who had access or chose to respond. Selection and coverage can remain related to the outcome, so population inference needs a defensible frame and selection route.
Does random assignment make a study representative?
Random assignment strengthens causal comparison among recruited units by balancing alternatives in expectation. Representation is a separate sampling question. A narrow volunteer experiment can have strong internal validity and limited external validity at the same time.
Assessment move
Learn measurement levels by their accumulated properties rather than by example lists. For every scale, say what a difference or ratio means. Draw population, frame, selected sample and respondent layers, then diagnose undercoverage and nonresponse. Compare probability techniques by their selection mechanism and non-probability techniques by the bounded purpose they can serve.
For experiments, write the treatment and control observations, calculate the intended contrast and name the specific pathway through which each validity threat could imitate an effect. Create a scale-classification set using unfamiliar variables rather than memorised examples. For each, state whether values provide identity, order, equal distance and a true zero, then name one permissible and one forbidden comparison.
Redesign weak rating questions by balancing categories, clarifying anchors, separating neutrality from uncertainty and checking that higher values always mean the stated direction. For comparative tasks, name the reference set and explain why a first rank does not establish absolute appeal. Rehearse sampling as four layers: target population, frame, selected sample and achieved respondents.
For each scenario, define elements, sampling units, extent and time. Mark undercoverage, ineligible entries, duplicates and nonresponse at the correct transition. Compare simple random, systematic, stratified and cluster sampling through their mechanism and consequences. For convenience, judgement, quota and referral routes, write a bounded purpose and the population claim that must be avoided.
Do not use sample size as a synonym for representativeness. For experiments, draw treatment and control lanes with observations, assignment and the intended contrast. State whether randomisation is assignment or sampling. Diagnose history, maturation, testing, instrumentation, selection, attrition and regression toward the mean by explaining the specific pathway through which each could create the observed difference.
Then separate internal from external validity across people, setting, time and implementation. Finish by writing the treatment as delivered, the counterfactual, the eligible population, the outcome window and the strongest causal conclusion. Add implementation, compliance and contamination checks without excluding cases merely because they weaken the contrast.
End with a three-column ledger for measurement, representation and causality. Place every planned claim in the column it requires and name the scale, sampling route or experiment that supports it. If one strength is being used to hide another weakness, repair the earlier layer. This habit prevents precise ratings from becoming population estimates and random assignment from being mistaken for representative recruitment.
Working through Questionnaires, Scaling, Sampling, and Experiments in MKF2121? Sia is AskSia’s AI Marketing tutor — ask any MKF2121 Questionnaires, Scaling, Sampling, and Experiments question and get a clear, step-by-step explanation grounded in how MKF2121 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
